Activity theory and distributed cognition: or What does cscw need to do with theories?

The setting is a hotline group (here abbreviated HLG) for personnel concerns at large company. HLG takes calls from both inside and outside the company. This particular call is an "employment verification”, where a caller (for example, a mortgage lender) contacts HLG to find out if a person is actually an employee. In order to answer this request the agent, Joan, must look up the person in a specific database, the EMPLOY system. Because of technical incompatibilities, the database must be accessed on a terminal separate from the one on her desk. This terminal (with EMPLOY) is shared by all of the agents, and it is located about three meters from Joan's desk. The agent, then, must disconnect her headset from the phone, walk to this central table and look up the person on the EMPLOY system. Furthermore, part of the HLG agent’s job is to maintain a record of call requests. To do this they use another computational system, the Call Tracking system (CAT), which is accessed from their desktop system.

The analysis began with the observations – primarily videotape, supported by additional direct observations, semi-structured interviews, social network analysis, and field notes. For the DCog analysis, the unit of analysis, as I discussed above, was clearly circumscribed by the extent of the phone call—because the temporal extent happened to coincide with the functional extent of the verification. Transcribing the call included actions as well as discourse. For privacy reasons we could only record half of the conversation, so we are limited in what we can directly observe. Table 1 shows the first three turns of this call in the transcript, interleaved with Joan’s actions.

Like AT there are many levels at which we can represent this. At the most basic level we detail:

1. 
   Representational states and the media they are instantiated in (or on);
   
2. 
   The character of the processing (such as creation, propagation, transformation) and a description of its mechanism
   
3. 
   Agents which enable the processing, whether human or artifact
   

At this stage all agents involved in the processing are enumerated, and only later pruned. Table 2 shows one detailed representation of the first three boxes from figure 1, which coincides with most of turn 3 (shown in Table 1). Reading down under each spoken fragment we see that the representational state is propagated through a variety of media and agents. A trio of entries details the agent, the representational state and medium, and the kind of processing. So Joan moving a mouse is represented by signifying the agent as Joan. The representational state is her hand position on the medium of the mouse. This is creating a physical process. The representational media detailed in the table are coordinated with each other to move the representational states through the processing necessary to accomplish the cognitive functioning of the system.

< Table 2 about here>

Notice that people and artifacts are treated equally as agents in some cases because they do processing. So Joan’s use of the mouse to drive the cursor to close a call tracking record and open another is represented as the propagation of her physical action, transformed through the mouse and the CPU that results in the cursor movement that appears on the CRT (first column Table 2). There is internal processing in Joan, and similarly there is implied internal processing that is happening inside the mouse and the computer CPU. In this representation internal processing has been left out. We generally ignore internal processing for two reasons: it is not the focus of the functional system, nor is it observable. But we often know that it is there. We can infer hidden from observable processes. In the case of computers we often have other means to know the internal processing, such as manuals. (Unfortunately we do not have the definitive manual on human processing.)

Figure 2 collapses some of the detail from the third column in this system into a diagrammatic “short hand” that re-represents it. Each box shows the agent, the representational state, and the media it is instantiated on. Joan says, “I just need to get, to get a little more information”. In saying this she does some internal processing that creates a representational state of the words carried on a vocal medium (i.e. her voice). This representational state is propagated verbally to the telephone, which then does its own processing, propagating the same representational state to the listener. At this level we presume the same medium. The caller does auditory processing on the same representational state.

One thing this figure points out is the problem of representing representational states and processes sufficiently. Table 2 is more explicit about what the processes are, while Figure 2 provides a better sense of the movement of the representational state as it gets processed. Figure 3 however gives a better idea of where memory is and foreshadows the result presented in Figure 1. It also gives a better representation of how agents bring representational states into coordination with each other to accomplish processing. Figure 3 uses yet another representation where agents, representational states, processes, and memory are all present. Agents are circular. Triangles represent memory. (The grayed-out triangle ‘switch’ as part of the telephone is unused memory.) Arrows represent the character of the processing; in this case more of a memory aid for the analyst to reconstruct what happened.

Thus, the caller says the social security number, which is processed through the telephone and via Joan’s auditory processing. Through rehearsal (saying each number) we infer that it is processed in her working memory, while we can observe her coordinating the audible representational state with her physical actions typing the same numbers into the Call Tracking (CAT) record. We again see Joan coordinating representational states as she reads from the CAT record and writes the number on a piece of paper. The paper is a special case because it is both a medium for the representational state and the “memory” of that state.

After this segment Joan writes down the social security number on a piece of paper and carries it over to the EMPLOY database to look up the information. If you compare Figure 3 with Figure 1 you can see that the analytic representation has been further focused to just one aspect of interest—in this case memory. Thus in figure 1 we track the propagation of what the caller says through the mediation of the phone’s memory, Joan’s and the caller’s short term and working memory, the Call tracking record, a piece of paper, and the EMPLOY database.

If this is confusing it is because these diagrams are private for the analyst; quite unlike the diagrams in AT which serve a more public and rhetorical purpose. You may notice that in figure 1 the telephone is a ‘memory’ while in figure 3 it has a component ‘switch’ memory. I have deliberately excluded some observed representations and processes related to the telephone system, both for clarity and because at this point in the analysis I am not yet concerned with it. If I care about the processing going on in the phone, I can make that its own unit of analysis or expand the detail of this analysis to include it. In fact, the phone turns out to be part of a sophisticated switch and record keeping process that propagates though the larger system of the hotline group (Ackerman and Halverson 1999). This is why in figure 1 it is part of the collection of memories that are the processing substrate for the system. Understanding the telephone’s role came from the extended ethnographic work—collecting information about how the phone was used. In turn, examining how the phone contributed to the propagation of those analyses was in fact fruitful.

As the range of presentations in the figures show there is not only one useful representation. Other representations, such as those used in Holder (Holder 1999) only indicate on what media the representational states appear at each time segment. This highlights which representational media are brought into coordination at which times. This is particularly useful in the case of an aircraft cockpit where most of the representational state appears on the medium of fixed instruments. Focusing in this way elides some constructs and highlights others. Part of what the ethnographic work contributes is to our understanding where such a representation will provide analytical power.

Because of the simplicity of this example one might imagine design solutions that would eliminate the need for Joan to do this kind of physical coordination of representational media. Electronically linking the CAT record system with the EMPLOY database on the same two computer systems could remove Joan from most of her coordinating role. One might imagine a voice response unit could eliminate her intervention entirely, while providing problems of its own (which would require seeing the results of another analysis). Seeing these alternative solutions is possible I argue, because the diagram is at the level of representational states.

This contrasts with how artifacts are treated as whole objects in AT. For example, Collins et al. adopt Engeström’s refinement of Wartofsky’s mediating artifacts hierarchy (Engeström, Miettinen et al. 1999) as a framework for both analysis and enumeration based on use. This makes it clear that, for example, conversation was a mediating “what” artifact, which contributes a means of achieving the object of knowledge authoring. “…hallway conversations and computer mediated conversations were central mediating artifacts, especially in transforming implicit knowledge to explicit and documented knowledge.” (p7) From this description we get little information at a level of detail that is useful for altering the design. That kind of information is exposed later when discussing the relationship between two elements of the activity system.

AT is an approach that analyzes the dialectical aspects of the activity system. Thus we learn more about how computer mediated conversation works within this activity system when the authors detail the tensions⁸ between the subject and mediating artifact elements. One example they give is about “InstaNote” a broadcast request and response tool. The authors argue that a subject’s description of the use of this artifact, in comparison to another artifact the chat window, implicitly provides design requirements. They quote an interviewee who states that InstaNote is more noticeable than the chat room tool because that tool is a little window that can get hidden. In contrast, InstaNote pops up a big yellow screen that cannot be missed. The implication is that changing notification on the chat room tool for similar obviousness to InstaNote would be good.

While interviews report the salient information, notice that it is a representation (representational state and its interpretation) that is the critical difference here. Direct observation and analysis with DCog, with its direct focus on representations might have suggested the same result. Further, it would also help examine the possibility of such a change, even if the interviewee had neglected to provide this information. Again, for both AT and DCog part of the usefulness of the approach relies on good ethnographic work to have exposed the data for observation. I think DCog makes the implication of the specific representations and the processes that coordinate them more evident. While communicating that is difficult because there are not named categories to put the observations in, the low level nature of analysis exposes the necessary information more directly.

3.5 The role and evidence of process

Although a DCog analysis is centered on cognitive processing, AT keeps process explicitly in the foreground by diagramming relations between elements within the activity system. In the analysis above we see how the framework of that diagram provides a way to walk through the various elements and their relationships to explain the workings of the system. By using it as a conceptual framework all interactions between subject and mediating artifact are considered together. Similarly, there is an assumed interaction between subject and object, which is mediated by artifacts. Stating it thus is an oversimplification of course, but again this is an example of how naming theoretical constructs helps AT both describe and communicate.

In a DCog analysis the mediating role of an artifact is not assumed between an object and subject, although it may occur that way. The telephone is a mediating artifact, but so is the content of the conversation over the telephone⁹. Breaking down an interaction into its respective representational states and examining the processes helps us understand more clearly what those representational states are doing, but it requires more work to clearly identify and talk about broad classes of processes. The process implicit in the interposition of artifacts mediating between subject and object is not assumed in DCog. The primacy of the processing of representational states for the cognitive work to get done and a theoretical language that is common to both individuals and artifacts, places all the processing at low level. Any one artifact may just help propagate a representation along a processing path, or it may play a different role with its own internal processing. After the low-level DCog analysis, and during the process of reconstructing a narrative of the cognitive work, a collection of phenomena might be labeled with the phrase ‘division of labor’ or ‘mediating artifact’. But what motivates the analysis, critique, and any subsequent design choices is not a separation of phenomena into named categories, but an analysis of the specifics of coordination of representational media.

Unfortunately, this low level focus may mean the reader overlooks the higher-level process. AT’s basic structure posits certain kinds of process interrelationships, which are implicit even when the analyst may not make them explicit. DCog obscures those relationships somewhat by focusing on the lower level. Where higher level phenomena have been drawn out of an analysis, such as Hutchins analysis of learning on both an individual and organizational level (Hutchins, 1995, Ch 6 & 7) constructs are usually not named and propagated through the theory. In addition, the lack of theoretical names makes it harder to bring it up to the higher level, requiring more descriptive work on the part of the analyst to explain the process at a higher level.